AMzhanghan / WDSR

A Pytorch implement of the paper 2018 NTIRE No.1 paper 《Wide Activation for Efficient and Accurate Image Super-Resolution》

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

2018.12视频通信大作业

A Pytorch implement of NTIRE2018 No.1 network WDSR https://arxiv.org/abs/1808.08718v1
Dataset: DIV2K 2017 https://data.vision.ee.ethz.ch/cvl/DIV2K/

DATA 
├── HR  
└── LR

Training data is augmented with random horizontal filp and rotations, check utility.py and rewrite class SRdataset!

How to train

Delete & make new

vim ./loss.log
mkdir ./samples
mkdir ./checkpoint

GPUs are needed for training

python main.py --cuda

How to test

Test method

700x700 HR image and its LR counterpart are randomly cropped from every image in DIV2K Validset
Calculate the mean PSNR of HR image and Image Restored by network

make correspond empty folder to store samples before test

mkdir ./foldername/

change samples save_path and model to restore in psnr.py

python psnr.py

Specific description of given samples, checkpoint as well as test results can be found in .numbers file ^_^

Result

Truth

LR

Output

About

A Pytorch implement of the paper 2018 NTIRE No.1 paper 《Wide Activation for Efficient and Accurate Image Super-Resolution》


Languages

Language:Python 100.0%